feat(pipeline): support configurable GPU architectures

[tetsuo-cpp]

Summary

• add a validated warpforthc --arch option and forward it to the NVVM target
• expose the supported architecture allow-list through pipeline validation and GPU test configuration
• constrain Vast.ai GPU selection to the requested compute capability and document supported targets

Verification

• cmake --build build
• cmake --build build --target check-warpforth (116 tests)
• uv run pytest -q -m "not gpu" (10 tests)
• cmake --build build --target check-clang-tidy
• cmake --build build --target check-format
• uv run ruff check gpu_test/
• uv run ruff format --check gpu_test/

Closes #63